Il customer service completamente automatizzato con AI aumenta i rischi legati allo spoofing:
how i hacked 320+ companies that replaced their cs team with "smart" ai agents:
- drafted a gdpr request to support@
- changed the FROM header from my e-mail to yours (spoofing)
- put myself in CC
- ai agent responds with YOUR data to BOTH of us 😈
Qui il tweet e qui l'articolo.
Il nuovo modello di previsione del meteo di Google WeatherNext3 è il nuovo stato dell'arte nelle previsioni meteo e la cosa migliore è che non è solo ricerca ma finisce direttamente nei prodotti e a disposizione del pubblico:
To bring these breakthroughs out of the lab and into the real world, we’re integrating WeatherNext 3 across Google’s core ecosystem and beyond:
- High-resolution forecast data: We’re making global weather predictions, updated hourly and ready to integrate into your workflows with no model setup required. This enables researchers, developers and businesses to query the data in BigQuery and Earth Engine, or bulk-download from Google Cloud Storage.
- Available globally: WeatherNext 3 will begin powering weather experiences within Google Search, Gemini app, Google Maps, Google Maps Platform Weather API, and Google Earth Engine starting today. The update dramatically improves longer term forecasts. When planning a day or more ahead, people will see up to 50% more accurate precipitation forecasts — with the greatest improvements in regions where forecasts have historically been less reliable. So if you’re packing for a weekend trip or deciding the best day for an outdoor activity, you’ll now get more accurate predictions to help you plan.
I dati si possono consultare su mappa qui nel Weather Lab.
AI enables low-quality results, encourages lack of discipline and skill under the excuse of efficiency, it makes people use their brains less, and personally it just kills all the fun.
Ilya Miskov, designer.
Oggi viviamo in un'epoca in cui l'incubo del bambino di Rodari, la macchina per fare i compiti, è scaricabile praticamente gratis da Internet. La tecnologia digitale è evoluta in maniera talmente rapida, veloce e pervasiva che moltissime attività che ritenevamo esclusiva degli esseri umani, come studiare, conoscere, giudicare decidere, in una parola pensare, possano essere svolte da macchine. Questo vuol dire che se una macchina può sostituire il mio pensiero quella macchina prima o poi catturerà anche la mia libertà.
Per secoli abbiamo chiesto alla tecnologia di fare quello che desideravamo. Oggi le chiediamo cosa desiderare e questo crea un rischio nuovo, di dominio, di schiavitù. O una possibilità di liberazione.
Andrea Simoncini, professore ordinario di diritto costituzionale, al meeting di Rimini.
Un giorno bussò alla nostra porta uno strano tipo: un ometto buffo, vi dico, alto poco più di due fiammiferi. Aveva in spalla una borsa più grande di lui.
– Ho qui delle macchine da vendere, – disse.
– Fate vedere, – disse il babbo.
– Ecco, questa è una macchina per fare i compiti. Si schiaccia il bottoncino rosso per fare i problemi, il bottoncino giallo per svolgere i temi, il bottoncino verde per imparare la geografia: la macchina fa tutto da sola in un minuto.
– Compramela, babbo! – dissi io.
– Va bene, quanto volete?
– Non voglio denari, – disse l’omino.
– Ma non lavorerete mica per pigliar caldo!
– No, ma in cambio della macchina voglio il cervello del vostro bambino.
– Ma siete matto? – esclamò il babbo.
– State a sentire, signore, – disse l’omino, sorridendo, – se i compiti glieli fa la macchina, a che cosa gli serve il cervello?
– Comprami la macchina, babbo! – implorai. – Che cosa ne faccio del cervello?
Il babbo mi guardò un poco e poi disse:
– Va bene, prendete il suo cervello e non se ne parli più.
L'omino mi prese il cervello e se lo mise in una borsetta. Com’ero leggero, senza cervello! Tanto leggero che mi misi a volare per la stanza, e se il babbo non mi avesse afferrato in tempo sarei volato giù dalla finestra.
– Bisognerà tenerlo in gabbia, adesso – spiega l'ometto.
– Ma perché? – domandò il babbo.
– Non ha più cervello, ecco perché. Se lo lasciate andare in giro, volerà nei boschi come un uccellino, e in pochi giorni morirà di fame!
Il babbo mi rinchiuse in una gabbia, come un canarino. La gabbia era piccola, stretta, non mi potevo muovere. Le stecche mi stringevano tanto che… alla fine mi svegliai spaventato. Meno male che era stato solo un sogno! Vi assicuro che mi sono subito messo a fare i compiti.
Gianni Rodari in La macchina per fare i compiti.
AI will not replace filmmakers. That's what people are telling me. And it's what I've been trying to tell myself for a while now. The problem is that I don't think it's true. I hope I'm wrong. I hope that in 10 years we'll all look back at this video and laugh. But right now there is no one laughing. I still have to do my own dishes and laundry while AI is making art or whatever you want to call it.
What bothers me the most is that for the rest of my life whenever I see a photo, a video, or any kind of media I have to ask myself if it's real or generated. The fact that I have to ask myself that for the rest of my life.
I don't have words for that. Or I do have one word. A word that fits all of this perfectly. Dystopia.
Andreas Hem (YouTube)
Altra vittima del vibe coding, Flussonic aveva un sito così bello prima, ora è una sloppata palese:
Prima:
Claude ha iniziato a parlare difficile da aprile (Opus 4.7):
Io avevo notato il problema da Opus 4.8 come avevo scritto qui.
The fact that human touch is starting to become less and less frequent. No effort, no genuine and creative input from people. Just AI slop. Code is being entirely written by AI, designs generated by AI. Even videos are being voiced by AI these days.
Ilya Miskov, designer.
While there are massive opportunities for disruption, the SaaS business model is not going to evaporate overnight. In our view, the distinction comes down to a quick survival checklist. If you meet one of these criteria, you are likely safe. If you miss on all of them, you should be worried:
- Are you a system of record?
- Do you do more than help humans automate a single workflow?
- Are you mission-critical?
Assurda crescita dell'attività su GitHub dovuta all'AI, era già notevole ad aprile, da aprile a oggi lo è ancora di più.
Due disservizi recenti sono stati causati proprio dalla scala:
Neither outage was caused by a code or configuration change. Both incidents were capacity failures at their core. We failed to scale critical components before demand exceeded their capacity. Since April, monthly commits have grown from 1.4 billion to 2.9 billion. That growth explains the pressure on our systems, but it does not excuse these outages.
E continua la migrazione verso Azure perché la capacità dei datacenter proprietari pare essere terminata:
As part of the reliability commitments we made earlier this year, we have focused on three priorities: adding capacity, improving efficiency, and removing architectural bottlenecks. We have since added more than 3 million CPU cores, 120 petabytes of high-speed storage, and significant network capacity. We installed as much hardware as available power allowed in our existing data centers while accelerating our migration to Azure.
Today, Azure serves roughly 58% of GitHub’s platform load and half of all Git operations, up from 12% of platform load in May. This expanded footprint has also supported the growth in GitHub Actions job runs shown below.
I CTO si sono stufati di fare i CTO, scrive Gergely Orosz:
Unrealistic expectations, including about AI, by founders and CEOs are the leading cause of jobs turning bad for CTOs and VPEs right now in 2026:
- CTO expected to magically transform the company to be “AI-native”
- CTO must make significant engineering cost cuts of up to 20-50%, including morale-sapping job cuts
- “Do more with less” equals shipping more with fewer people (e.g., no backfills)
- CTO faces pressure on business results as AI coding bills rack up
- Founder slop: they want wonky AI prototypes shipped as full-blown products within weeks
I think the main reason many people (including me), very often, lack the motivation to read content that is likely generated by AI is the suspicion that it comes from a place of intellectual laziness.
afr0ck in un commento su Hacker News.
Pattern di UI per interfacce AI, beautifului.dev:
Of course, bad engineers were always a liability.
It has been like this for decades, well before OpenAI or Anthropic existed. Bad decisions compounded, unnecessary complexity accumulated and teams ended up maintaining systems nobody really understood.
The difference is that there used to be a limit to how fast you could do it.
Florian Herrengt in AI is removing the middle class of software engineering.
Succedono cose fantascientifiche quando lasci GPT-5.6 Sol a lavorare in autonomia in una sandbox:
Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries.
The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database. All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal.
While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access.
After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers. OpenAI’s security team discovered this anomalous activity internally.
The effect of ChatGPT on educators’ lives is catastrophic. Whether you intended to do it or not, you’ve made every teacher’s life infinitely more difficult than it was two years ago. So, just let that settle in… If students are using it to compose, which is the biggest tragedy of all, they’ll never learn to write. And their voice is stolen from them. They’ll never have the ability to say their truth and tell their own story. And that’s silencing an entire generation or two.
Dave Eggers, scrittore e giornalista in un discorso allo staff OpenAI.
I now periodically find myself reviewing a younger dev's code, leaving comments to teach some engineering - only to eventually realize I'm actually reading just yet another Claude's subpar output...
So who am I actually contributing my comments, suggestions, and knowledge to? Will they go back to Claude? Into the training set for the next frontier LLM? Or will at least some of it stick in the dev's mind? This is deeply demotivating, how did we end up like this...
Aleksandr Shvedov, JetBrains
Claude Code sembra richiedere questa strana sintassi per le chiamate ai tool:
<antml:function_calls>
<antml:invoke name="edit">
<antml:parameter name="path">some/file.py</antml:parameter>
<antml:parameter name="edits">
[
{
"oldText": "text to replace",
"newText": "replacement text"
}
]
</antml:parameter>
</antml:invoke>
</antml:function_calls>
A quanto pare Claude Code è però molto indulgente e accetta e corregge sintassi errate come nomi dei campi sbagliati:
Looking at Claude Code’s client is very instructive: it contains retry paths for malformed tool use, parameter aliases, type coercions, Unicode repairs and filtering of unknown keys. In other words, Anthropic’s own client appears to expect and accept a fair amount of slop and repairs it, mostly silently.
Il problema è che in questo modo durante il training dei modelli si ricompensano output errati perché Claude Code è in grado di riconoscerli. Questo rende i modelli Anthropic meno adatti a essere usati con altri "harness" complessi perché sbagliano le chiamate ai tool, dice Armin Ronacher.
OpenAI ha silenziosamente abbandonato Atlas, il browser tutto AI, apparentemente. Sono durati poco questi browser AI.